Context-dependent role of digital rectal examination in a contemporary prostate biopsy cohort: differential predictive value with and without multiparametric magnetic resonance imaging

Abstract

Objective

To quantify the context-dependent contribution of digital rectal examination (DRE) to the prediction of clinically significant prostate cancer (csPCa, International Society of Urological Pathology [ISUP] grade group ≥2) in a contemporary biopsy cohort, comparing a nonmultiparametric magnetic resonance imaging (mpMRI) clinical pathway with an mpMRI-informed pathway including prostate imaging–reporting and data system (PI-RADS).

Materials and methods

We performed a retrospective analysis of 666 consecutive prostate biopsies at a single academic center (July 2017–December 2024). DRE was performed by urology residents and attending urologists. Two multivariable logistic regression models were constructed: Model A (full cohort, n = 612) included age, log-prostate-specific antigen (PSA), log-PSA density (PSAD), prior biopsy, and DRE; Model B (mpMRI subcohort, N = 306) additionally included PI-RADS score. Discrimination was assessed using area under the curve (AUC) and likelihood ratio tests. Sensitivity analyses evaluated any cancer (ISUP ≥1) and high-grade cancer (ISUP ≥4).

Results

csPCa prevalence was 30.7% (202/660), and DRE was suspicious in 52.3% (346/662). In Model A, DRE was independently associated with csPCa (adjusted odds ratio [aOR], 2.18, 95% confidence interval [CI]: 1.45–3.28, P < 0.001), although the gain in discrimination was modest (ΔAUC = 0.009). In Model B, DRE showed borderline significance (aOR, 1.88, 95% CI: 1.01–3.52, P = 0.048) with negligible discrimination gain (ΔAUC = 0.002). Sensitivity analyses showed consistent findings. A dose–response gradient was observed: csPCa rates were 19.2% (normal), 31.6% (cT2a), and 58.9% (cT2b–T4) (p-trend <0.0001).

Conclusions

DRE was independently associated with csPCa in the absence of mpMRI, but its incremental discriminative contribution was modest and became negligible once PI-RADS was incorporated. These findings support a limited, context-dependent role for DRE in contemporary biopsy pathways.

1. Introduction

Digital rectal examination (DRE) has been a cornerstone of prostate cancer (PCa) detection since the pre-prostate-specific antigen (PSA) era, yet its role in the contemporary diagnostic pathway remains debated.^1,2 The adoption of multiparametric magnetic resonance imaging (mpMRI) and targeted biopsy protocols has shifted the diagnostic paradigm toward imaging-driven pathways,^3,4 raising the question of whether DRE provides clinically meaningful incremental value when advanced imaging data are available.

Prior studies have addressed this question with mixed results.5, 6, 7 Large screening cohorts have consistently shown low sensitivity and specificity for DRE when used alone,^1,8 while biopsy-based series report higher diagnostic performance due to verification bias.^9,10 The critical clinical question is not whether DRE detects cancer in isolation, but whether it adds information beyond what PSA, PSA density (PSAD), and prostate imaging–reporting and data system (PI-RADS) already provide in the prebiopsy risk assessment.^11,12

Additionally, in training settings, DRE competency acquisition by urology residents represents an important educational outcome.¹³ The degree to which resident-performed DRE achieves diagnostic accuracy comparable to that of expert examiners informs curricular design and supervision requirements.

The objectives of this study were: (1) to quantify how the predictive contribution of DRE changes across two contemporary diagnostic settings—a non-mpMRI pathway and an mpMRI-informed pathway including PI-RADS—and to assess the extent to which PI-RADS subsumes the information conveyed by DRE; (2) to evaluate whether DRE retains a dose-response relationship with clinically significant prostate cancer through DRE-derived clinical T-stage categories; and (3) to explore examiner-level performance in the subset with available training data.

2. Materials and methods

2.1. Study design and population

This was a retrospective cohort study of consecutive patients undergoing prostate biopsy at Instituto Nacional de Ciencias Médicas y Nutrición "Salvador Zubirán" between July 2017 and December 2024. All patients with a clinical indication for prostate biopsy (elevated PSA, suspicious DRE, or imaging findings) were eligible. Patients with incomplete pathology data were excluded (n = 6), yielding 660 evaluable cases from a total of 666 registered procedures. The study was approved by the institutional ethics committee (URO-5623-25-26-1).

2.2. Clinical variables

DRE was performed prior to biopsy by the examining urologist (resident or attending). DRE findings were recorded using institutional clinical examination categories. G1–G4 denoted graded enlargement patterns based on estimated prostate size on DRE, broadly corresponding to benign prostatic enlargement, whereas cT2a indicated a unilateral palpable nodule and cT2b–cT4 indicated progressively more suspicious or advanced palpable findings. For the main binary analysis, G1–G4 categories were grouped as nonsuspicious (normal) DRE findings, while cT2a–cT4 findings were classified as suspicious. PSA and prostate volume were obtained at the time of biopsy indication. PSA density was calculated as PSA divided by prostate volume (mL). Patients who underwent mpMRI had PI-RADS v2.1 scores³ assigned by dedicated uroradiologists. The decision to perform mpMRI was based on clinical judgment and imaging availability, both of which evolved over the study period as mpMRI became progressively incorporated into the institutional prebiopsy pathway.

2.3. Biopsy technique

All biopsies were performed as transrectal ultrasound-guided systematic biopsies (12 cores). Patients with mpMRI-detected lesions (PI-RADS ≥3) additionally underwent cognitive or software-assisted fusion targeted biopsies.¹⁴ Transperineal biopsies were performed in 10 cases (1.5%) based on clinical indication.

2.4. Outcome definition

The primary outcome was clinically significant prostate cancer (csPCa), defined as International Society of Urological Pathology (ISUP) grade group ≥2 on biopsy pathology.¹⁵ Secondary outcomes included any prostate cancer detection (ISUP ≥1) and high-grade cancer (ISUP ≥4).

2.5. Statistical analysis

Two multivariable logistic regression models were prespecified.¹⁶ Model A included the full evaluable cohort (complete cases for age, log-PSA, log-PSAD, prior biopsy, and DRE) without imaging variables. Model B was restricted to patients with mpMRI and included PI-RADS score as an additional covariate. Fusion biopsy modality was excluded from both models due to conceptual redundancy with PI-RADS (the decision to perform fusion is a consequence of PI-RADS score). PSA and PSAD were log-transformed to normalize their distributions. Collinearity was assessed by variance inflation factors with centered variables.¹⁷

Discrimination was evaluated by the area under the receiver operating characteristic curve (AUC).¹⁸ The incremental value of DRE was assessed by comparing the AUC of models with and without DRE and by likelihood ratio tests. Calibration was assessed by the Hosmer–Lemeshow test.¹⁶ DRE diagnostic performance (sensitivity, specificity, predictive values, likelihood ratios) was calculated for the overall cohort and stratified by resident training level where examiner data were available.

Prespecified sensitivity analyses repeated both models using (1) any prostate cancer (ISUP ≥1) and (2) high-grade cancer (ISUP ≥4) as outcomes to assess whether DRE retained predictive value across the spectrum of disease severity. An additional analysis evaluated DRE as a 3-level variable (normal, cT2a, cT2b+) to assess a dose-response relationship. Missing data patterns were characterized by comparing included and excluded patients on key variables, with statistical tests (Mann–Whitney U for continuous, chi-square for categorical variables) to assess whether missingness was consistent with a missing-completely-at-random mechanism.¹⁹

All analyses were performed in Python 3.12 using statsmodels 0.14, scikit-learn 1.5, and SciPy 1.14. Two-sided P < 0.05 was considered statistically significant.

3. Results

3.1. Cohort characteristics

A total of 666 biopsy procedures were registered. After excluding 6 patients with incomplete pathology, 660 were evaluable. Median age was 68 years (interquartile range [IQR], 63–73), median PSA was 8.6 ng/mL (IQR, 5.8–14.6), and median PSAD was 0.16 ng/mL² (IQR, 0.10–0.27). DRE was suspicious in 346/662 (52.3%) and normal in 316/662 (47.7%). A total of 321 patients (48.3%) underwent mpMRI prior to biopsy, and 191 (28.7%) received fusion-targeted biopsies. Prior biopsy had been performed in 180 patients (27.2%). Any prostate cancer was detected in 323/660 (48.9%), csPCa (ISUP ≥2) in 202/660 (30.6%), and high-grade cancer (ISUP ≥4) in 91/660 (13.8%). Table 1 presents the full cohort characteristics. Patient selection and analytic cohorts are shown in Supplementary Fig. 1.

Table 1.

Cohort characteristics (n = 666 procedures, 660 evaluable)

Variable	Value	n available
Age, years, median (IQR)	68 (63–73)	660
PSA, ng/mL, median (IQR)	8.6 (5.8–14.6)	658
PSAD, ng/mL,2 median (IQR)	0.16 (0.10–0.27)	617
DRE suspicious, n (%)	346 (52.3%)	662
mpMRI performed, n (%)	321 (48.3%)	659
PI-RADS 1–2	116 (36.1%)	321
PI-RADS 3	56 (17.4%)	321
PI-RADS 4	95 (29.6%)	321
PI-RADS 5	54 (16.8%)	321
Fusion biopsy, n (%)	191 (28.7%)	660
Prior biopsy, n (%)	180 (27.3%)	659
Any PCa (ISUP ≥1), n (%)	323 (48.9%)	660
csPCa (ISUP ≥2), n (%)	202 (30.6%)	660
High-grade (ISUP ≥4), n (%)	91 (13.8%)	660

csPCa, clinically significant prostate cancer; DRE, digital rectal examination; IQR, interquartile range; mpMRI, multiparametric magnetic resonance imaging; PI-RADS, prostate imaging–reporting and data system; PSA, prostate-specific antigen; PSAD, PSA density.

3.2. DRE diagnostic performance

Overall DRE sensitivity for csPCa was 70.3% (142/202), specificity 55.7% (254/456), positive predictive value (PPV) 41.3%, and negative predictive value (NPV) 80.9%. The positive likelihood ratio was 1.59 and the negative likelihood ratio was 0.53.²⁰ At lower prevalences than the biopsy cohort (30.7%), NPV increases (e.g., 94.4% at 10% prevalence) while PPV decreases (15.0% at 10% prevalence), consistent with expected Bayesian behavior.

3.3. Multivariable analysis: Model A (full cohort, N = 612)

In the complete-case analysis without PI-RADS (N = 612, Table 2), DRE was independently associated with csPCa (adjusted odds ratio [aOR], 2.18, 95% confidence interval [CI]: 1.45–3.28, P < 0.001). Log-PSAD was the strongest predictor (aOR, 3.18, 95% CI: 2.12–4.76, P < 0.001).¹¹ Age contributed modestly (aOR, 1.03 per year, P = 0.008). Log-PSA was nonsignificant after adjustment for PSAD (aOR, 0.82, P = 0.36), consistent with the known collinearity between PSA and PSAD. The model AUC was 0.798 with DRE versus 0.788 without (ΔAUC = 0.009). The likelihood ratio test for DRE addition was highly significant (χ² = 14.3, P < 0.001). Hosmer–Lemeshow goodness-of-fit was adequate (χ² = 9.2, P = 0.33).¹⁶

Table 2.

Multivariable logistic regression for csPCa (ISUP ≥2): Model A (full cohort without PI-RADS, n = 612)

Variable	aOR	95% CI	P
Age (per year)	1.03	1.01–1.06	0.008
log (PSA)	0.82	0.54–1.25	0.363
log (PSAD)	3.18	2.12–4.76	<0.001
Prior biopsy	1.17	0.75–1.83	0.488
DRE (suspicious)	2.18	1.45–3.28	<0.001

AUC without DRE: 0.788; AUC with DRE: 0.798; ΔAUC: 0.009; LR test χ² = 14.3, p < 0.001. Hosmer–Lemeshow χ² = 9.2, P = 0.33.

aOR, adjusted odds ratio; CI, confidence interval; csPCa, clinically significant prostate cancer; DRE, digital rectal examination; PSA, prostate-specific antigen.

3.4. Multivariable analysis: Model B (mpMRI subcohort, n = 306)

Among 306 patients with complete data including PI-RADS (Table 3), DRE was borderline significant (aOR, 1.88, 95% CI: 1.01–3.52, P = 0.048). PI-RADS was the dominant predictor (aOR, 2.13 per category, p < 0.001),^3,21 followed by log-PSAD (aOR, 3.40, P < 0.001). The addition of DRE to the model increased AUC from 0.804 to 0.806 (ΔAUC = 0.002, LR test χ² = 3.9, P = 0.047). Variance inflation factor values were all below 2.5 (range 1.01–2.44), indicating acceptable collinearity.¹⁷

Table 3.

Multivariable logistic regression for csPCa (ISUP ≥2): Model B (mpMRI subcohort with PI-RADS, n = 306)

Variable	aOR	95% CI	P
Age (per year)	1.00	0.96–1.04	0.916
log (PSA)	1.02	0.51–2.02	0.957
log (PSAD)	3.40	1.67–6.93	<0.001
PI-RADS (per category)	2.13	1.57–2.89	<0.001
Prior biopsy	1.17	0.63–2.17	0.622
DRE (suspicious)	1.88	1.01–3.52	0.048

AUC without DRE: 0.804; AUC with DRE: 0.806; ΔAUC: 0.002; LR test χ² = 3.9, p = 0.047. Hosmer–Lemeshow χ² = 14.0, P = 0.08.

aOR, adjusted odds ratio; CI, confidence interval; csPCa, clinically significant prostate cancer; DRE, digital rectal examination; PI-RADS, prostate imaging–reporting and data system; PSA, prostate-specific antigen; PSAD, PSA density.

The multivariable associations for Models A and B are summarized in Fig. 1. Receiver operating characteristic curves comparing models with and without DRE are shown in Fig. 2.

Fig. 1.

Multivariable predictors of clinically significant prostate cancer. Forest plot showing adjusted odds ratios (aOR) from multivariable logistic regression models predicting clinically significant prostate cancer (ISUP ≥2). Model A includes the full cohort without PI-RADS, and Model B includes the mpMRI subcohort with PI-RADS. Suspicious DRE was independently associated with csPCa in Model A and borderline significant in Model B.

csPCa, clinically significant prostate cancer; DRE, digital rectal examination; mpMRI, multiparametric magnetic resonance imaging; PI-RADS, prostate imaging–reporting and data system.

Fig. 2.

Incremental discrimination of DRE in predictive models. Receiver operating characteristic (ROC) curves comparing models with and without DRE for predicting clinically significant prostate cancer in the full cohort (Model A) and the mpMRI subcohort (Model B). The addition of DRE produced minimal improvement in discrimination when PI-RADS was included.

DRE, digital rectal examination; mpMRI, multiparametric magnetic resonance imaging; PI-RADS, prostate imaging–reporting and data system.

3.5. DRE dose-response by clinical T-stage

When DRE was categorized into three levels based on clinical T-stage (Supplementary Table 1), a clear dose-response gradient was observed for csPCa detection: 19.2% for normal DRE, 31.6% for cT2a, and 58.9% for cT2b–T4 (χ² for trend = 69.3, P < 0.0001). In Model B with 3-level DRE, the cT2a category was independently significant versus normal (aOR, 2.24, 95% CI: 1.12–4.48, P = 0.022), while cT2b+ was not significant (aOR, 1.82, 95% CI: 0.74–4.48, P = 0.196), likely because PI-RADS captures the same information for advanced palpable disease. The dose–response relationship between DRE clinical T-stage and cancer detection is illustrated in Fig. 3.

Fig. 3.

Dose–response relationship between DRE findings and cancer detection. Cancer detection rates according to DRE clinical T-stage categories. Panel A shows the prevalence of any prostate cancer and clinically significant cancer across normal, cT2a, and cT2b–T4 findings. Panel B demonstrates a strong dose-response gradient for csPCa with increasing DRE abnormality.

csPCa, clinically significant prostate cancer; DRE, digital rectal examination.

3.6. Sensitivity analyses

To assess robustness across the disease severity spectrum, we repeated both models using different outcome definitions (Supplementary Table 2). For any prostate cancer (ISUP ≥1, prevalence 48.9%), DRE remained significant in Model A (aOR, 1.43, 95% CI: 1.00–2.04, P = 0.050) but not in Model B (aOR, 1.33, 95% CI: 0.80–2.19, P = 0.27). For high-grade cancer (ISUP ≥4, prevalence 13.8%), DRE was significantly associated in Model A (aOR, 2.31, 95% CI: 1.28–4.16, P = 0.006) and borderline significant in Model B (aOR, 2.71, 95% CI: 0.98–7.49, P = 0.055). The consistent pattern—DRE significant in Model A but attenuated in Model B—was observed across all outcome definitions, confirming that PI-RADS subsumes most of the information provided by DRE.

3.7. Missing data assessment

Patients excluded from Model B (those without mpMRI, n = 360) differed significantly from included patients (n = 306) on several key variables (Supplementary Table 3). Excluded patients had higher PSAD (P = 0.004), a higher rate of suspicious DRE (58.4% vs. 45.1%, P < 0.001), higher csPCa prevalence (36.2% vs. 24.2%, P = 0.001), and fewer prior biopsies (14.7% vs. 41.8%, P < 0.001). Age and PSA did not differ significantly. This pattern is consistent with missing-at-random conditional on era, as mpMRI was adopted progressively over the study period and earlier patients with higher-risk profiles were less likely to have imaging.¹⁹ Characteristics of patients included vs. excluded from the mpMRI subcohort are shown in Supplementary Fig. 2.

3.8. Resident DRE performance (subset analysis, n = 204)

Examiner training level was documented in 206/666 cases (30.9%), yielding 204 cases with complete DRE and pathology data. Post-graduate year-1 (PGY-1) residents performed 76 examinations, PGY-2 residents performed 115, and PGY-3 residents performed 13. In this documented subset, crude DRE performance estimates were similar between PGY-1 and PGY-2 residents (sensitivity 65.4% vs. 65.7%, specificity 62.0% vs. 66.2%, NPV 77.5% vs. 81.5%; Supplementary Table 4).¹³ PGY-3 performance is reported descriptively only because of the very small sample size (n = 13). Because examiner data were missing for 69.1% of the cohort, these findings should be interpreted as exploratory and not as definitive evidence of training-level differences. Diagnostic performance by resident training level is presented in Supplementary Fig. 3.

4. Discussion

This study should be interpreted not as a redemonstration that DRE improves contemporary MRI-based risk stratification, but as a quantification of how its contribution changes according to diagnostic context. In our cohort, DRE remained independently associated with csPCa in the non-mpMRI setting, although the corresponding improvement in discrimination was modest (Model A: aOR, 2.18, P < 0.001; ΔAUC = 0.009). Once PI-RADS was incorporated, the additional discriminative contribution of DRE became negligible (Model B: ΔAUC = 0.002), despite borderline statistical significance. Sensitivity analyses across outcome definitions showed the same directional pattern, supporting the view that much of the information conveyed by DRE is subsumed by mpMRI in imaging-based pathways, while DRE may still retain contextual clinical information where MRI is unavailable or not routinely performed. This pattern suggests that, in contemporary practice, DRE functions less as an independent discriminator within MRI-based pathways and more as a contextual or staging-oriented examination when imaging is unavailable.

The independent association observed in Model A is consistent with prior biopsy-cohort studies reporting odds ratios between 1.5 and 3.0 for suspicious DRE and csPCa.^5,6,9 Our finding of aOR 2.18 falls within this expected range and was supported by an adequate sample size (n = 612), acceptable model calibration (Hosmer–Lemeshow P = 0.33), and a significant likelihood ratio test (P < 0.001). However, in our data, this association translated into only a modest gain in discrimination, underscoring that statistical significance does not necessarily imply major incremental clinical utility. In Model B, the combination of borderline significance (P = 0.048) and near-null discrimination gain (ΔAUC = 0.002) strongly suggests that PI-RADS captures most of the information conveyed by DRE for biopsy decision-making, which is biologically plausible because both assessments reflect, through different modalities, underlying tumor burden and local extent.^21,22

A novel contribution of this study is the demonstration of a clear dose-response gradient using granular DRE data: the csPCa rate increased from 19.2% for normal DRE to 31.6% for cT2a to 58.9% for cT2b–T4 (p-trend <0.0001).²³ This gradient was maintained in the adjusted analysis for cT2a (aOR 2.24, P = 0.022) but not for cT2b+ (aOR, 1.82, P = 0.196), suggesting that the most advanced palpable disease is already captured by PI-RADS in the mpMRI subcohort. Nonetheless, the descriptive dose-response supports the continued role of DRE for clinical staging, a function distinct from its role in biopsy-decision models.

The prevalence-adjusted analysis provides important context for interpreting DRE predictive values.²⁰ The NPV of 80.9% observed in this biopsy cohort (prevalence 30.7%) would be 94.4% at 10% prevalence (primary care referral) and 97.3% at 5% prevalence (screening). Conversely, the PPV drops to 15.0% and 7.7% at these lower prevalences. This means DRE has reasonable negative predictive utility in lower-risk populations but poor PPV, supporting its role as a rule-out rather than a rule-in tool when used without imaging.

Resident-level performance data were available for only 204 of 666 cases (30.9%), and within this documented subset, crude sensitivity and specificity estimates appeared similar between PGY-1 and PGY-2 residents.¹³ However, the substantial missingness, the very small PGY-3 sample, and the absence of examiner identifiers or temporal adjustment preclude any firm conclusion regarding training-level differences or competency acquisition. The apparent similarity between PGY-1 and PGY-2 performance may reflect either early skill acquisition or confounding by case selection. Without individual examiner identifiers, repeated examinations, or interobserver agreement data, these results should be interpreted as exploratory and descriptive rather than as evidence of stable examiner-level reproducibility.

4.1. Comparison with prior literature

Our findings align with the emerging consensus that DRE adds limited incremental value to imaging-based diagnostic pathways.24, 25, 26, 27 The PRECISION and MRI-FIRST studies established the central role of mpMRI-based risk stratification in contemporary biopsy pathways.^4,28 In that context, our Model B results, including only negligible discrimination gain after adding DRE (ΔAUC = 0.002), suggest that PI-RADS captures much of the clinically relevant information previously conveyed by DRE for biopsy decision-making. However, our Model A results indicate that DRE may still retain practical value where mpMRI is unavailable or not routinely performed, a situation that remains common in many healthcare systems.²⁹ A key contribution of the present study is that both scenarios were evaluated within the same contemporary institutional cohort using parallel multivariable models, allowing a direct estimate of how the apparent value of DRE changes once PI-RADS enters the diagnostic framework. From a practical standpoint, these findings suggest that DRE is most useful when mpMRI is unavailable, delayed, or not routinely integrated into the prebiopsy pathway, and as a bedside component of clinical staging. By contrast, when high-quality mpMRI with PI-RADS scoring is already available, DRE is unlikely to materially improve biopsy decision models, although it may still contribute to the overall clinical examination. In this context, contemporary pathways may be better refined by MRI-complementary tools rather than by reintroducing DRE into imaging-informed prediction models.³⁰

4.2. Limitations

Several limitations should be considered when interpreting these findings. First, this was a retrospective single-center biopsy cohort, which limits the transportability of the observed associations to other practice settings. Because all patients underwent biopsy regardless of DRE result, verification bias is also present, and the true specificity and NPV of DRE cannot be estimated for the broader population of men with elevated PSA.³¹ Second, the mpMRI subcohort (Model B, n = 306) was not randomly selected and differed from the non-mpMRI group in several clinically relevant characteristics, including PSAD, suspicious DRE frequency, csPCa prevalence, and prior biopsy status. In addition, mpMRI was adopted progressively over the study period, introducing a potential calendar-time effect and temporal confounding in the comparison between imaging and nonimaging pathways. This introduces a risk of selection bias and means that the attenuation of DRE in Model B should be interpreted cautiously, as part of the observed effect may reflect subgroup composition rather than only the intrinsic dominance of PI-RADS. The missing-data mechanism for mpMRI availability was compatible with missing-at-random conditional on era, but this assumption remains unverifiable.¹⁹ Third, examiner training level was missing for 69.1% of the cohort, substantially limiting the resident-level analysis, which should therefore be regarded as exploratory. Fourth, because DRE is inherently operator-dependent, the absence of individual examiner identifiers and repeated assessments precluded clustering or interobserver agreement analyses.³² This limits our ability to determine whether part of the observed DRE effect reflects true clinical signal versus variability in examination technique or threshold for classifying an examination as suspicious. Finally, the study was not powered to detect very small AUC differences, so the near-null discrimination gain observed in Model B should be interpreted primarily as evidence of limited incremental utility rather than as a definitive estimate of zero effect.³³

5. Conclusions

DRE has a limited, context-dependent role in contemporary biopsy pathways. In the absence of mpMRI, it remained independently associated with csPCa and preserved a dose-response relationship across clinical T-stage categories, although the corresponding gain in discrimination was modest. Once PI-RADS was incorporated, its incremental discriminative contribution became negligible, suggesting that MRI captures much of the information conveyed by DRE for biopsy decision-making. These findings support retaining DRE as a pragmatic, bedside- and staging-oriented tool where mpMRI is unavailable or not routinely incorporated, while recognizing that it is unlikely to materially improve MRI-informed biopsy decision models.

Conflict of interest

The authors declare no conflicts of interest relevant to this work.

Appendix A. Supplementary data

The following are the Supplementary data to this article: